The precise mass dependence of respiratory O2 consumption underpins the “oxygen triple-isotope” approach to quantifying gross primary productivity in modern and ancient environments. Yet, the physical-chemical origins of the key 18O/16O and 17O/16O covariations observed during respiration have not been tied to theory; thus, the approach remains empirical. First-principles calculations on enzyme active-site models suggest that changes in the O–O bond strength upon electron transfer strongly influence respiratory isotopic fractionation. However, molecular diffusion may also be important. Here, we use measurements of the relative abundances of rare isotopologues 17O18O and 18O18O as additional tracers of mass dependence during dark respiration experiments of lacustrine water. We then compare the experimental results to first-principles calculations of O2 interacting with heme-oxidase analogues. We find a significantly steeper mass dependence, supported by theory, than has been previously observed. Enrichments of 17O18O and 18O18O in the O2 residue suggest that θ values are strongly influenced by chemical processes, rather than being dominated by physical processes (i.e., by bond alteration rather than diffusion). In contrast, earlier data are inconsistent with theory, implying that analytical artifacts may have biased those results. Implications for quantifying primary productivity are discussed.